Introduction
Chickenpox is an acute infectious disease caused by the varicella-zoster virus. Children and adolescents are highly susceptible to chickenpox infection, and it is characterized by maculopapular rashes on the skin and mucous membranes with mild systemic symptoms. Chickenpox has a high incidence but low mortality rate, and it is one of the most common childhood diseases [1]. Although most cases have mild clinical symptoms, if not treated promptly, the infection may lead to postherpetic neuralgia, which affects the quality of life and can lead to death in severe cases [2].
Chickenpox is prevalent worldwide, and it has a seasonal pattern. The European region has a single peak pattern from March to May, whereas Asian countries have a double peak pattern from March to May and from December to January [3]. According to the World Health Organization, it is estimated that there are approximately 4.2 million hospitalizations and 4200 deaths due to serious complications of varicella annually worldwide, and the danger of varicella has been seriously underestimated [4]. Despite widespread immunization, varicella continues to spread and develop in many countries, such as the United States, Italy, and various European countries, due to its high transmissibility [3-6]. In China, according to an epidemiological survey of varicella in China from 2005-2019, a total of 6,442,147 cases of varicella were reported nationwide. The reported incidence of varicella increased from 41,211 cases (3.17/100,000) in 2005 to 7,979,482 cases (70.14/100,000) in 2019. The average annual incidence of chickenpox showed an increasing trend each year from 2005 to 2019, and the number of reported cases of varicella increased nearly 22 times from 2005 to 2019. Chickenpox now has one of the highest incidences in terms of preventable diseases in China [7].
Yunnan Province has a high incidence of chickenpox in China, and it ranked among the top 5 provinces in the country in 2019 in terms of the incidence in the population aged under 14 years, accounting for 76.36% of all chickenpox cases. Chickenpox public health emergencies mainly occur in rural elementary schools, and the situation is serious as the chickenpox epidemic continues to grow year by year. However, attention to the management and elimination of chickenpox public health emergencies in rural elementary schools and kindergartens is still insufficient, and the problem of untimely reporting of chickenpox incidence remains. Yunnan Province needs to take effective measures to suppress the spread and prevalence of chickenpox [8].
Currently, problems such as low chickenpox vaccine coverage, the breakthrough of the chickenpox vaccine, the lack of attention to chickenpox disease, imperfect chickenpox surveillance reports, and the lack of public information on chickenpox epidemics are still present in China [9]. In China, chickenpox is reported through the Chinese Disease Prevention and Control Information System. Because chickenpox is a viral disease with mild and self-limiting symptoms, it has not been included in the Chinese National Disease Reporting System for statutory infectious diseases. There is no uniform standard for reporting chickenpox cases across China. However, the public health emergency management information system has focused on some aggregated outbreaks [9,10], but some disseminated cases or subclinical infections may be overlooked.
At present, the means of chickenpox prevention and control in China are relatively singular and limited to the traditional reported incidence monitoring and isolation treatment of patients with chickenpox [2]. The actual incidence of chickenpox in most provinces is not publicly available, and the public cannot obtain timely information on the chickenpox epidemic. Apart from public health departments, the public does not have effective channels to obtain real-time information on local chickenpox incidence, which is not conducive to taking active self-protective measures against chickenpox.
The internet has become an increasingly popular means of accessing health information. The increase in web-based information provides a potentially useful source of data for disease surveillance. Due to its real-time nature and ease of access, internet data can be used to fill the gaps in traditional public health surveillance [11]. Therefore, the possibility of using internet data for the surveillance of various diseases is being increasingly explored as an aid to improve disease prevention and control. The use of internet search data as a complementary means to traditional infectious disease surveillance methods is particularly applicable to neglected diseases that are less affected by consultations [12].
Previous studies [12,13] have indicated that Google Trends can well demonstrate the epidemic characteristics of chickenpox abroad and is suitable for the simulation of features of infectious diseases in epidemic cycles. Moreover, Bakker et al [12] studied chickenpox incidence patterns using Google Trends and found that Google search data show a high correlation (R2=0.65-0.71) with varicella outbreaks. Thus, Google Trends can be used for chickenpox incidence prediction and early warning. In China, more studies have focused on vaccine efficacy analysis and descriptive studies of chickenpox prevalence [7-10], and most of the exploration of chickenpox surveillance issues has been performed by using actual incidence report data to build models to predict the onset of chickenpox [2], whereas there are few studies exploring whether the Baidu search engine can be applied to chickenpox epidemic surveillance and early warning.
Google Trends is not highly used in China. Google search data do not reflect the true search tendencies of the Chinese public because Baidu is the most widely used search engine product in China. The Baidu index (BDI) [14], which was established based on Baidu search information, should reflect the search needs and awareness of internet users as well as Google Trends abroad in China [15,16]. A considerable number of studies have been conducted using the BDI for China-wide disease prediction. For example, one study has successfully predicted the epidemic trend of influenza using the BDI [17]. Similarly, some surveillance studies using BDIs for diseases, such as brucellosis; dengue fever; and hand, foot, and mouth disease, have shown that BDIs can be used to reflect disease prevalence [18-20].
The current chickenpox surveillance system in China is flawed and incomplete. The flawed surveillance system and the lack of public information on chickenpox outbreaks have led to a weak response to chickenpox outbreaks in China. Public health events caused by chickenpox epidemics have persisted for a long time, generating a large number of patients with chickenpox, which disrupts the study, work, and life of patients while also affecting the normal production and everyday life of families, schools, and workplaces. Moreover, chickenpox epidemics impose a heavy burden on the healthy socioeconomic development of China.
This study focused on the following specific research questions:
What are the correlations between BDI scores and actual varicella incidence data in Yunnan Province?
Can internet data be used to predict future varicella disease epidemics?
Can big data be used as a supplement to traditional surveillance systems for early warning surveillance of infectious diseases as well as epidemics?
Discussion
Principal Findings
In this study, we used the chickenpox BDI to build a model to predict the incidence of chickenpox during the same period. We concluded that it is feasible to use BDI data for varicella epidemic surveillance, which does not rely on actual reported data of varicella incidence.
The Principle of Using the BDI to Predict Incidence
The BDI reflects changes in public interest in various events over a period of time. As a relatively neglected disease, news reports about chickenpox are substantially fewer than those about other infectious diseases, such as influenza and AIDS. Possible reasons for public interest in chickenpox are news reports or illnesses in their own family, suggesting that the main source of web-based search data may come from patients with chickenpox and their family members. This hypothesis is the reason why we believe that the chickenpox BDI can reflect the dynamics of chickenpox incidence through search information.
The use of internet data for disease surveillance has several advantages. For instance, it is more cost-effective and less time-consuming compared to traditional surveillance methods that rely on patient consultations and medical records [21,22]. Moreover, internet data can capture disease trends earlier than traditional surveillance methods because they are updated in real time. In addition, they can detect disease outbreaks that may go unnoticed using traditional surveillance methods. As neglected diseases often go unnoticed due to the limited health care resources and infrastructure in low-income countries, internet data can serve as a useful tool for monitoring these diseases. By leveraging internet data, health officials can detect disease outbreaks earlier, allocate resources more effectively, and ultimately prevent the spread of diseases more efficiently. Therefore, the use of internet data for disease surveillance has the potential to revolutionize disease prevention and control efforts.
Based on this principle, we believe that the BDI can be used as a new infectious disease surveillance tool to complement traditional public health surveillance systems. By combining big data epidemiological surveillance methods with traditional epidemiological surveillance tools, governments can establish a more sensitive surveillance tool. Real-time and low-cost internet data can help government health departments quickly identify diseases, populations, and areas with potential transmission risks and take effective measures.
Comparison With Prior Work
This study explores the potential of using internet data for disease surveillance, with a focus on chickenpox. Although previous research has shown success in using Google for disease prediction [21-23], we wanted to investigate whether Baidu, a search engine with higher usage rates in China, could also be used for this purpose. Our findings confirm that Baidu can be used for disease monitoring, providing an opportunity to better leverage internet data to predict disease occurrence and its spread. This discovery is particularly important for public health institutions, which can now make more accurate predictions and take more targeted preventive measures. Furthermore, this study highlights the growing importance of internet information as a source of data for disease surveillance in the future.
In this study, the chickenpox BDI was highly consistent with the actual occurrence trend of chickenpox (r=0.747), and compared to the disease studies conducted using the BDI [15,17-20,24-27], the correlation coefficients of these studies were generally between r=0.3 and r=0.93, indicating that the chickenpox BDI well reflects the actual trend of chickenpox occurrence. Thus, it is feasible to use the BDI for chickenpox epidemic surveillance. At the same time, studies exploring the correlation between Google Trends and chickenpox [12] have shown that the average correlation between chickenpox and Google Trends is approximately r=0.762 globally, indicating that the use of the BDI for chickenpox in China has similar surveillance effects as those obtained using Google Trends abroad. Therefore, in China, the methods and ideas of using Google for varicella incidence trend prediction in foreign countries can be applied to the BDI to conduct supplementary surveillance of chickenpox incidence.
In this study, the SVR model outperformed the multiple linear regression model both in terms of fitting and prediction (Tables 2 and 3), which showed a mean absolute percentage error of 9.1%, RMSE of 189, and R2 of 0.548. Comparing it with previous studies [19,28], the SVR model outperformed the autoregressive model for sexually transmitted diseases and the autoregressive integrated moving average model for brucellosis in terms of mean absolute percentage error and RMSE. Overall, the SVR model showed promising results for predicting chickenpox incidence and can be further improved and compared with other models for infectious diseases in future studies.
The SVR model developed using the BDI accurately predicted the actual number of current varicella cases. Unlike previous chickenpox prediction models, the method developed in this study provided a rapid assessment of the current chickenpox epidemic without relying on actual incidence reporting data. The method allows for immediate calculation of chickenpox case prediction, which is more rapid than traditional prediction systems, allowing it to be used as a rapid method to help the public know the current chickenpox dynamics. Moreover, due to the simplicity of the method and easy access to data sources, it is likely to be applicable to most places in China. When model predictions increase, it may indicate a rise in the number of chickenpox cases, allowing for disease control and prevention–related authorities to prepare for potential chickenpox outbreaks.
Limitations
This study had several limitations that deserve further discussion. First, China is a vast country with regional differences in customs and culture as well as population distribution and geographic ethnicity [29]. Thus, it is not possible to predict the incidence trend in all provinces using one model. If data can be collected for each region, the prediction of chickenpox incidence trends can be tailored to different regions. Second, similar to other studies exploring the relationship between disease and the internet [23], when the media reports the chickenpox epidemic, people without chickenpox may search for chickenpox out of curiosity and fear, which would lead to a surge in the number of searches. Thus, further research is needed to eliminate the influence of the media on the results. Finally, multiple regression modeling based on BDI data alone does not allow for accurate prediction of chickenpox dynamics, and the volume of disease search data does not correspond one to one with the number of reported cases. Infectious disease prediction models based solely on internet search data may also be confounded by searchers’ knowledge of the disease and local language restrictions.
Therefore, in future studies, we should also consider additional influencing factors related to chickenpox outbreaks, such as climate, economy, and vaccination status, for a comprehensive analysis to achieve accurate prediction of chickenpox incidence.
Conclusions
Based on the results, it is feasible to apply the BDI method to reflect the incidence of varicella. The fitted and predicted values of the SVR model were consistent with the actual incidence trend of chickenpox, indicating that the model based on the BDI can be used to reflect the actual local incidence trend of chickenpox in real time. Thus, the BDI can be used for disease surveillance. Internet search data can be used as a supplement to traditional surveillance systems to help with the early detection of potential disease outbreaks or disease epidemics.